報告題目：Multi-task Deep Learning for Human Action Parsing
報告人：邵嶺(Ling Shao) 博士, Chair Professor, Fellow of IET
單位：Northumbria University
報告時間：2016年9月1日(周四)下午14:30-16:00
報告地點：屯溪路校區(qū)逸夫樓408會議室
報告摘要：Automatically recognizing objects, scenes and actions is a core component of an artificial intelligence system. In this talk, I will cover two of my main research areas – human action recognition and deep learning. Action recognition has been an active research topic in computer vision due to its various applications in human-machine interaction, robotics, video surveillance and visual big data search. I will first review some related work on handcrafted features, feature/deep learning and attributes learning. Then I will introduce our recent multi-task system that can jointly solve three main problems: 1) Where in the video do the actions occur? (2) What categories do the actions belong to? and (3) How are these actions performed? This multi-task learning framework is designed based on a state-of-the-art 3D deep convolutional neural network (3D-CNN). Specifically, in the training phase, action localization, classification and attributes learning can be jointly optimized via the proposed deep architecture. Once model training is completed, given an upcoming test video, we can describe each individual action in the video simultaneously as: where the action occurs, what the action is and how the action is performed. To train the deep network, we also introduce a new large-scale aligned action dataset, NASA, with 200K well labeled video clips. Finally, I will present the results of detailed action parsing on challenging, realistic datasets that are collected by us or publicly available. Some initial results on zero-shot learning via the obtained action attributes will be discussed too.
報告人簡介：邵嶺教授于中國科學(xué)技術(shù)大學(xué)獲得學(xué)士學(xué)位，在英國牛津大學(xué)獲得碩士和博士學(xué)位。邵嶺教授在學(xué)術(shù)界和工業(yè)界有著豐富的工作經(jīng)驗，現(xiàn)任英國Northumbria University 計算機與信息科學(xué)系首席教授，并將于2016年11月份加盟University of East Anglia任計算機系首席教授。邵嶺教授在計算機視覺、圖像/視頻處理、模式識別與機器學(xué)習(xí)領(lǐng)域取得了豐碩的研究成果，在PAMI、TIP、TNNLS、IJCV等頂尖國際期刊和CVPR、ICCV、ECCV、MM、IJCAI等頂尖國際會議發(fā)表論文超過200篇，并擁有超過10項美國/歐盟專利。邵嶺教授目前擔(dān)任TNNLS、TIP、TCSVT、ToC等頂尖期刊的副主編，同時也是ICPR、BMVC、ICME等著名國際會議領(lǐng)域主席，ICCV、CVPR、ECCV、MM等頂尖會議程序委員會委員。邵嶺教授是IET學(xué)會會士、英國計算機學(xué)會會士、IEEE高級會員、ACM終身會員。
太陽集團tyc5997
TOP
